TzuMao Litzumao@berkeley.edu

I am a postdoc at the EECS department of UC Berkeley, working with Jonathan RaganKelley. My work is at the intersection of computer graphics, computational photography, and programming systems. Specifically, I develop efficient sampling methods for light transport simulation, and programming systems that extract domain knowledge from graphics and image processing algorithms (through, for example, automatic differentiation). I did my Ph.D. in the computer graphics group at MIT CSAIL, advised by Frédo Durand. I received my B.S. and M.S. degree in computer science and information engineering from National Taiwan University in 2011 and 2013, respectively. During my time at National Taiwan University, I was a member of the graphics group at Communication and Multimedia Lab, where I worked with YungYu Chuang.
Learning to Optimize Halide with Tree Search and Random Programs
Andrew Adams, Karima Ma, Luke Anderson, Riyadh Baghdadi, TzuMao Li, Michaël Gharbi, Benoit Steiner, Steven Johnson, Kayvon Fatahalian, Frédo Durand, Jonathan RaganKelley ACM Transactions on Graphics (Proceedings of SIGGRAPH 2019) The first Halide autoscheduler that produces faster code comparing to human experts on average. 

Samplebased Monte Carlo Denoising using a KernelSplatting Network
Michaël Gharbi, TzuMao Li, Miika Aittala, Jaakko Lehtinen, Frédo Durand ACM Transactions on Graphics (Proceedings of SIGGRAPH 2019). Permutation invariant mapping from Monte Carlo samples to an image through splatting. 

Differentiable Visual Computing [slides (Keynote)] [slides (Powerpoint)]
TzuMao Li MIT PhD Dissertation A coherent view of my PhD research. It has some new discussions regarding previous papers, and some background reviews. 

Inverse Path Tracing for Joint Material and Lighting Estimation
Dejan Azinović, TzuMao Li, Anton Kaplanyan, Matthias Nießner Conference on Computer Vision and Pattern Recognition (CVPR), 2019 (oral presentation) Applying differentiable rendering for material and lighting reconstruction. 

Differentiable Monte Carlo Ray Tracing through Edge Sampling
TzuMao Li, Miika Aittala, Frédo Durand, Jaakko Lehtinen ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia 2018) Computing gradients of the light transport equation through an explicit sampling of Dirac delta functions on triangle edges. 

Differentiable Programming for Image Processing and Deep Learning in Halide
TzuMao Li, Michaël Gharbi, Andrew Adams, Frédo Durand, Jonathan RaganKelley ACM Transactions on Graphics (Proceedings of SIGGRAPH 2018) Halide meets automatic differentiation. 

Aether: An Embedded Domain Specific Sampling Language for Monte Carlo Rendering
Luke Anderson, TzuMao Li, Jaakko Lehtinen, Frédo Durand ACM Transactions on Graphics (Proceedings of SIGGRAPH 2017) A programming language for Monte Carlo rendering that automatically computes the probability density of a light path sample. 

Anisotropic Gaussian Mutations for Metropolis Light Transport through HessianHamiltonian Dynamics
TzuMao Li, Jaakko Lehtinen, Ravi Ramamoorthi, Wenzel Jakob, Frédo Durand ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia 2015) A variant of Metropolis light transport algorithm that makes use of automatically differentiated Hessian matrix of light path contribution. 

DualMatrix Sampling for Scalable Translucent Material Rendering
YuTing Wu, TzuMao Li, YuHsun Lin, and YungYu Chuang IEEE Transactions on Visualization and Computer Graphics (TVCG), 2015 Subsurface scattering with manylights using matrix sampling. 

SUREbased Optimization for Adaptive Sampling and Reconstruction
TzuMao Li, YuTing Wu, YungYu Chuang ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia 2012) Stein's unbiased risk estimator for sampling and denoising in Monte Carlo rendering. 
Graphics bibtex
A mega bibtex file containing many graphicsrelated literatures. 

Joint Stein’s Unbiased Risk Estimation for Adaptive Sampling and Reconstruction
A short note on a generalized formulation of our SUREbased rendering method. 

dpt
My prototypical renderer. 

smallgdpt
GradientDomain Path Tracing in ~450 lines. 